A Design Method of Two-Dimensional Subwavelength Grating Filter Based on Deep Learning Series Feedback Neural Network

Area of Science:

• Photonics and Optical Engineering
• Computational Physics
• Materials Science

Background:

• Subwavelength grating structures offer excellent filtering properties but traditional design methods are computationally intensive.
• Efficient design of these structures is crucial for advanced optical applications.

Purpose of the Study:

• To develop a novel, computationally efficient design method for two-dimensional (2D) subwavelength grating filters.
• To enable both forward simulation and backward design of grating parameters using a neural network approach.

Main Methods:

• A series feedback neural network was developed and programmed in Python.
• A dataset of 46,080 data points was generated using rigorous coupled-wave analysis (RCWA) simulations.
• The network considered parameters such as shape, height, period, duty cycle, and waveguide layer height.

Main Results:

• The optimal neural network achieved a low loss function of 0.024.
• The network successfully designed 2D subwavelength gratings within 1.12 seconds.
• A strong correlation (greater than 0.65) was found between the design results and theoretical spectra.

Conclusions:

• The proposed series feedback neural network offers a significantly quicker and more accurate alternative to traditional methods for designing 2D subwavelength gratings.
• This method addresses challenges in network reverse design convergence and provides a new pathway for optical filter development.